NI’m not sure why that’s a surprise, if you’re spawning 1000s of closures in any systems you’ll accumulate memory usage. Once the tasks all resolve and you’ve stopped creating new ones, does the memory settle again as the GC cleans up?
NWell the GC will try to wait until it really has to do work as it blocks the whole thread while it does. So if there’s memory available and you’re still asking for more tasks it will probably just wait
NI’m not saying there’s no chance of a memory leak, but the description you’ve given sounds like a great way to make any API appear to be leaking memory
NBuild your use case, this synthetic stuff isn’t particular helpful
NcreateContext is called for every invocation
NYou're microoptimising. 280mb for a node app with Prisma in is nothing at all.
NI'm telling you it's not a problem
NIf you want to use 28mb of RAM, use C++
NLike I said before, based on your description you are creating 100s of requests every second, and you resolve them after some number of seconds. Increasing memory usage shouldn't be a surprise here
NYou're holding a lot of stuff in memory while those requests fulfill
NNot necessrily
NThings get cached for legitimate reasons, going up to 280mb isn't really a shock
Nprisma alone I bet caches a lot of stuff
NHow do you know? Have you built it?
NUltimately this is not a tRPC question, this is a question of deploying some code on a RPi and needing to design a process which is memory efficient